JP4296290B2 - Speech recognition apparatus, speech recognition method and program - Google Patents

Description

  The present invention relates to a speech recognition device, a speech recognition method, and a program, and more particularly, to a speech recognition device, a speech recognition method, and a program for recognizing speech that is uttered in syllable units.

  Conventionally, various speech recognition apparatuses have been proposed in order to improve the voice input speed.

  In Patent Document 1, a syllable string corresponding to a syllable (for example, a syllable string “ai”) is registered as a standard pattern, and it is difficult to recognize a single syllable alone. Even if there is a voice, the syllable string “A” is recognized with a high probability if the syllable string “Ai Ueno A” is input.

  Patent Document 2 describes that speech is recognized with high accuracy by using both speech generated by dividing a word into syllables and speech generated continuously from the same word.

Patent Document 3 describes changing the facial expression of a character in accordance with the reliability level in order to make it easy to confirm the voice recognition result and the reliability level.
JP 9-179578 A JP 10-340096 A Japanese Patent Laid-Open No. 9-292895

  In Patent Document 1, in order to recognize a single syllable “A”, the user must generate a syllable string “Ai Ueno A” corresponding thereto. For this reason, there has been a problem that the voice input speed is slow.

  In Patent Document 2, the user has to utter the same word in two ways, which causes a problem that the voice input speed becomes slow.

  In Patent Document 3, only the reliability of the voice recognition result is fed back to the user, and this alone cannot increase the voice input speed.

  The present invention has been proposed to solve the above-described problems, and provides a speech recognition device, a speech recognition method, and a program that can improve the input speed of speech uttered by syllable. Objective.

The speech recognition device according to claim 1 is a speech input means for inputting speech uttered in syllable units, a syllable recognition means for recognizing speech input by the speech input means for each syllable, Output means for outputting the recognition result of the syllable in a manner corresponding to the reliability of the recognition result of the syllable recognition means , wherein the output means has a first threshold value indicating that the reliability of the recognition result is unrecognizable. When it is lower, it outputs predetermined information as the recognition result of the syllable, and when the reliability of the recognition result is equal to or higher than the first threshold and lower than the second threshold indicating incomplete recognition, At least a vowel of the syllable is output as a syllable recognition result .

According to a fourth aspect of the present invention, there is provided a speech recognition method comprising: a syllable recognition step for recognizing speech uttered in units of syllables for each syllable; and a syllable in a manner corresponding to a reliability of a recognition result obtained by the syllable recognition step. An output step of outputting the recognition result of the first step, and in the output step, when the reliability of the recognition result is lower than a first threshold value indicating that the recognition is impossible, information predetermined as the recognition result of the syllable is obtained. When the reliability of the recognition result is equal to or higher than the first threshold and lower than the second threshold indicating incomplete recognition, at least a vowel of the syllable is output as the recognition result of the syllable .

According to an eleventh aspect of the present invention, there is provided a speech recognition program, comprising: a voice input unit that inputs a voice uttered in units of syllables; and a syllable recognition that recognizes a voice input by the voice input unit for each syllable. And an output means for outputting the recognition result of the syllable in a manner corresponding to the reliability of the recognition result of the syllable recognition means, and the output means determines that the reliability of the recognition result is unrecognizable. When the value is lower than the first threshold value, information that is predetermined as the syllable recognition result is output, and the reliability of the recognition result is equal to or higher than the first threshold value and the second threshold value indicating incomplete recognition. When the value is low, the syllable recognition result is output at least as a vowel of the syllable.

  The syllable recognition means recognizes, for each syllable, the voice uttered by the user in syllable units. Here, there are various recognition results of the syllable recognition means, such as those with high reliability and those with low reliability. However, even when the reliability of the recognition result is low, if the user does not provide any information back to the user, the user often does not try to speak the next syllable.

  The output means outputs the syllable recognition result in a manner corresponding to the reliability of the recognition result of the syllable recognition means. That is, the output means changes the output mode of the recognition result according to the reliability of the recognition result of the syllable.

  Therefore, according to the above invention, the input of the next syllable can be prompted, and as a result, the voice input speed can be improved.

  When the reliability of the recognition result of the syllable is less than the first threshold, the syllable is not recognized at all. At this time, if the output means does not output any information about the syllable, the user does not try to speak the next syllable.

  Therefore, according to the above invention, when the reliability of the recognition result of the syllable is lower than the first threshold value indicating that the recognition is impossible, the syllable cannot be recognized at all by outputting the predetermined information as the recognition result of the syllable. Even in this case, the user can be prompted to utter the next syllable. Note that the predetermined information may be information such as a combination.

  When the reliability of the recognition result of the syllable is not less than the first threshold and less than the second threshold, only a part of the syllable is recognized, but not completely recognized. For example, vowels that are part of the syllable are recognized, but consonants that are other parts are not recognized.

  At this time, if the output means does not output any information about the syllable, the user does not try to speak the next syllable. On the other hand, when outputting some information, it is possible to prompt the user to speak by outputting only the recognized part.

  Therefore, according to the invention, when the reliability of the recognition result is equal to or higher than the first threshold and lower than the second threshold indicating incomplete recognition, by outputting at least the syllable vowel as the syllable recognition result, Since the user can be notified that a part of the syllable has been recognized, the user can be prompted to utter the next syllable.

The speech recognition apparatus according to claim 2 is the invention according to claim 1 , wherein the output means is at least one of a sound output means for outputting sound and an image output means for outputting an image. .

A speech recognition method according to a fifth aspect of the present invention is the voice recognition method according to the fourth aspect of the present invention, wherein at the output step, at least one of a voice and an image is output.

A speech recognition apparatus according to a third aspect of the invention is the invention according to the first or second aspect , wherein the syllable string candidate storage means stores a plurality of syllable string candidates, and the syllable string candidate storage means. Selecting means for selecting a syllable string candidate most corresponding to the syllable string composed of the plurality of syllable strings recognized by the syllable recognition means from among the plurality of syllable string candidates stored in the syllable recognition unit.

The speech recognition method according to claim 6 is the invention according to claim 4 or claim 5 , comprising a plurality of syllables recognized in the syllable recognition step from a plurality of syllable string candidates. And a selection step of selecting a syllable string candidate most corresponding to the syllable string.

  There is no point in recognizing only one syllable, and ultimately it is necessary to recognize a syllable string composed of a plurality of syllables. Here, syllable string candidates that are meaningful words, such as nouns, are prepared in advance. Then, the selecting means selects a syllable string candidate most corresponding to the syllable string composed of a plurality of already recognized syllables from the syllable string candidates.

  Thus, according to the above-described invention, a syllable string composed of a plurality of syllables that are uttered separately can be recognized with high accuracy and reliability even if a part of the syllable is unrecognizable or incompletely recognized.

  The speech recognition apparatus, the speech recognition method, and the program according to the present invention recognize the speech uttered in units of syllables for each syllable, and output the syllable recognition result in a manner according to the reliability of the recognition result. The input of the next syllable can be prompted, and as a result, the voice input speed can be improved.

  Hereinafter, the best mode for carrying out the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration of a speech recognition apparatus according to an embodiment of the present invention. The speech recognition device recognizes speech uttered by the user in syllable units. In the present embodiment, an example will be described in which the user utters “hi, g, shi, ya, ma, do, u, bu, tsu, e, n” in units of syllables.

  The speech recognition apparatus includes a microphone 1 that inputs speech uttered by a user and generates a speech signal, a speech segment extractor 2 that extracts a speech segment from the speech signal and extracts acoustic parameters, and single syllable speech recognition. A single syllable recognizer 3 to perform, a syllable candidate selector 4 for selecting a syllable string candidate and outputting a final recognition result, a syllable string candidate dictionary database 5 for storing a dictionary representing a plurality of syllable string candidates, A display device 6 that outputs the recognition result as an image and a speaker 7 that outputs the recognition result as sound are provided.

  The single syllable recognizer 3 performs speech recognition of a single syllable and calculates the reliability of the recognition result of the single syllable. A plurality of single syllables recognized by the single syllable recognizer 3 is referred to as a “recognition target syllable string”. The recognition target syllable string may include syllables that cannot be recognized or are incompletely recognized (only vowels can be recognized).

  The syllable string candidate dictionary database 5 stores a plurality of syllable string candidates and an acoustic model string corresponding to each syllable string candidate. In the present embodiment, the syllable string candidate is not particularly limited as long as it has a single meaning, but in this embodiment, it is assumed to be a noun such as a place name or a facility.

  The syllable candidate selector 4 selects a syllable string candidate from the syllable string candidate dictionary database 5 based on the recognition target syllable string, matches the acoustic parameter with the acoustic model string, and finally determines the recognition target syllable string. A simple speech recognition result.

  FIG. 2 is a flowchart showing a procedure of voice recognition processing by the voice recognition device. When the user generates a voice divided for each syllable, the microphone 1 converts the voice into a voice signal and supplies the voice signal to the voice segment extractor 2.

  The voice segment extractor 2 accepts the input of the voice signal supplied from the microphone 1 (step ST1), cuts the voice segment from the voice signal, and performs acoustic analysis to extract feature parameters (acoustic parameters) (step). ST2).

  The single syllable recognizer 3 recognizes single syllables using the acoustic parameters extracted by the voice segment extractor 2, and calculates the reliability of the recognition result (step ST3). At this time, the display device 6 and the speaker 7 output the recognition result in a manner corresponding to the reliability of the recognition result of the single syllable.

  Specifically, when the reliability of the recognition result is high, the display device 6 displays the single syllable character image as it is. When the reliability of the recognition result is slightly low (for example, when the reliability is equal to or higher than the first threshold and lower than the second threshold: only the vowel is recognized), the display device 6 displays the character image of the vowel. Next, a predetermined image “?” Indicating that consonant recognition is impossible is displayed. In addition, when the reliability of the recognition result is low (for example, when the reliability is less than the first threshold: both vowels and consonants cannot be recognized), the display device 6 cannot recognize the syllable and inputs the next syllable. A predetermined image “*” representing prompting is displayed.

  On the other hand, the speaker 7 outputs the synthesized speech of the single syllable when the reliability of the recognition result is high, and the synthesized speech of the vowel when the reliability of the recognition result is a little low (when only the vowel can be recognized). Only output. Further, when the reliability of the recognition result is low (when both vowels and consonants cannot be recognized), the speaker 7 outputs a synthesized speech “Yes” indicating that syllable recognition is impossible and prompting the next syllable input. .

  As described above, even if the reliability of the recognition result of the single syllable is not high, the speech recognition apparatus operates the system correctly by feeding back the reliability of the recognition result to the user through vision or hearing. Can be reminiscent of the user. As a result, the user can be prompted to input the next single syllable.

  Next, the single syllable recognizer 3 determines whether or not the recognition result of the single syllable is correct (step ST4). Here, the user can confirm the recognition result of the single syllable via the output of the display device 6 and the speaker 7. When the user determines that the recognition result is incorrect, for example, the user can press a button (not shown) indicating that the recognition result is incorrect.

  The single syllable recognizer 3 determines that the recognition result of the single syllable is correct when it does not detect the button press even after a predetermined time has elapsed, and recognizes the single syllable when the button press is detected. It is determined that the result is not correct. In step ST4, instead of making the user determine whether the answer is correct, for example, it may be determined whether the single syllable recognizer 3 is automatically correct according to the reliability of the single syllable.

  When the single syllable recognizer 3 determines that the recognition result of the single syllable is not correct, the single syllable recognizer 3 performs the syllable recognition again by outputting the recognition result of the single syllable which is the next candidate (step ST3). The single syllable recognizer 3 repeatedly executes the processing of step ST3 and step ST4 until the recognition result of the single syllable becomes correct.

  On the other hand, when the single syllable recognizer 3 determines that the recognition result is correct, the single syllable recognizer 3 stores the waveform data (acoustic parameter) of the single syllable in a memory (not shown) (step ST5).

  Next, the single syllable recognizer 3 determines whether or not all syllable inputs have been completed (step ST6). Here, the single syllable recognizer 3 determines that the syllable input is not completed when the next single syllable is input, and returns to step ST1. Further, if the next single syllable is not input even after a predetermined time has elapsed after the processing of step ST5, it is determined that the syllable input is completed.

  Thereby, for example, when the syllable recognition process for the single syllable “hi” is completed in steps ST1 to ST5, the process returns to step ST1 again, and the syllable recognition process for the next single syllable “ga” is performed. Then, the syllable recognition process is performed for each single syllable constituting the syllable string “hi-ga-shi-ya-ma-do-u-bu-tsu-e-n”.

  FIG. 3 is a diagram for explaining a recognition result for each single syllable of the input speech “Hi, Gashi, Ya, Ma, Do, U, Bu, Tsu, E, N”.

  Every time the user utters a single syllable, the display device 6 recognizes “*”, “? A”, “shi”, “? A”, “ma”, “do”, “u”, “?” As recognition results. U, “*”, “e”, and “*” images are sequentially output. At the same time, the speaker 7 recognizes “Yes”, “A”, “Shi”, “A”, “Ma”, “Do”, “U”, “U”, “Yes”, “E”, "Yes" synthesized speech is output sequentially.

  Thereby, the user can utter a single syllable continuously without being influenced by the reliability of the recognition result of the single syllable. That is, since the speech recognition apparatus prompts the user to continuously speak single syllables even if there is an error in part of the recognition result of single syllables, it is possible to avoid correction for each single syllable as much as possible. it can. A plurality of single syllables recognized as shown in FIG. 3 are referred to as “recognition target syllable strings”.

  The syllable candidate selector 4 matches the recognition target syllable string in the syllable string candidate dictionary stored in the syllable string candidate dictionary database 5 (for example, the same syllable / vowel having the same number of single syllables). A syllable string candidate that matches the condition such as being in the same position is selected (step ST7). The syllable candidate selector 4 regards an unrecognizable single syllable corresponding to “*” as an arbitrary character. Thereby, the syllable candidate selector 4 selects a syllable string candidate even if the recognition target syllable string includes a single syllable that could not be partially recognized.

  Further, the syllable candidate selector 4 reads out the acoustic parameters of each syllable from the memory, matches the acoustic parameters of the recognition target syllable string and the acoustic model of each syllable string candidate, and corresponds most to the recognition target syllable string. Select a candidate. Specifically, the score is calculated for the recognition target syllable string and each syllable string candidate, and the syllable string candidate with the highest score is selected as a re-syllable recognition result (final recognition result) (step ST8). Then, the display device 6 and the speaker 7 output the final recognition result.

  As described above, the speech recognition apparatus according to the embodiment of the present invention prompts the user to continuously speak single syllables even if the recognition result does not completely match the single syllable spoken by the user. Therefore, apparent misrecognition can be reduced and the voice input speed can be improved.

  Further, in the post-processing after step ST7, the speech recognition apparatus narrows down syllable string candidates from recognition target syllable strings including unrecognizable or incompletely recognized single syllables, and then repeats the sound of the same recognition target syllable string. Using parameters, syllable string candidates corresponding to recognition target syllable strings can be output as final recognition results. That is, by comparing the recognition target syllable string that is the recognition result with the syllable string candidates in the syllable string candidate dictionary database 5 and narrowing down the input speech, even if single syllable recognition is impossible or incomplete, The recognition target syllable string can be recognized with high accuracy.

  In addition, this invention is not limited to embodiment mentioned above, It can apply also about what was changed in design within the range described in the claim.

  For example, a speech recognition program for executing the processes from step ST1 to step ST8 described above is installed in a computer, and a speech segment extractor 2, a single syllable recognizer 3, a syllable candidate selector 4, a syllable string candidate are installed in the computer. The function of the dictionary database 5 may be executed. The computer may install a voice recognition program transmitted via a communication line, or may install a voice recognition program recorded on a recording medium such as an optical disk, a magnetic disk, or a semiconductor memory.

  Further, the display device 6 may display other symbols, characters, and characters instead of “?” Or “*”. Similarly, instead of “Yes”, the speaker 7 may output other synthesized speech such as “e”, for example, information such as “Kai”.

It is a block diagram which shows the structure of the speech recognition apparatus which concerns on embodiment of this invention. It is a flowchart which shows the procedure of the speech recognition process by a speech recognition apparatus. It is a figure for demonstrating a speech recognition result.

Explanation of symbols

1 Microphone 2 Voice segment extractor 3 Single syllable recognizer 4 Syllable candidate selector 5 Syllable string candidate dictionary database 6 Display device 7 Speaker

Claims (7)

A voice input means for inputting voice uttered in syllable units;
Syllable recognition means for recognizing the voice input by the voice input means for each syllable;
Output means for outputting the recognition result of the syllable in a manner according to the reliability of the recognition result of the syllable recognition means ,
The output means outputs information predetermined as the recognition result of the syllable when the reliability of the recognition result is lower than a first threshold indicating that recognition is impossible, and the reliability of the recognition result is the first A speech recognition device that outputs at least a vowel of the syllable as a recognition result of the syllable when the threshold is lower than a second threshold that is not less than the second threshold and indicates incomplete recognition. The output means is at least one of a sound output means for outputting sound and an image output means for outputting an image.
The speech recognition apparatus according to claim 1 . Syllable string candidate storage means for storing a plurality of syllable string candidates;
Selecting means for selecting a syllable string candidate most corresponding to a syllable string composed of a plurality of syllable strings recognized by the syllable recognition means from a plurality of syllable string candidates stored in the syllable string candidate storage means; The speech recognition apparatus according to claim 1, further comprising: A syllable recognition process for recognizing speech uttered in syllable units for each syllable;
And an output step of outputting a recognition result of the syllable in a manner that depends on the reliability of the recognition result of the syllable recognition step
In the output step, when the reliability of the recognition result is lower than a first threshold indicating that recognition is not possible, information predetermined as the recognition result of the syllable is output, and the reliability of the recognition result is the first reliability. A speech recognition method for outputting at least a vowel of the syllable as a recognition result of the syllable when the threshold is lower than a second threshold that indicates incomplete recognition. In the output step, at least one of sound and image is output.
The speech recognition method according to claim 4 . A selection step of selecting a syllable string candidate most corresponding to the syllable string composed of a plurality of syllable strings recognized in the syllable recognition process from a plurality of syllable string candidates.
The speech recognition method according to claim 4 or 5 . Computer
A voice input means for inputting voice uttered in syllable units;
Syllable recognition means for recognizing the voice input by the voice input means for each syllable;
Function as output means for outputting the recognition result of the syllable in a manner according to the reliability of the recognition result of the syllable recognition means ,
The output means outputs information predetermined as the recognition result of the syllable when the reliability of the recognition result is lower than a first threshold indicating that recognition is impossible, and the reliability of the recognition result is the first A speech recognition program that functions to output at least a vowel of the syllable as a recognition result of the syllable when the threshold is lower than a second threshold that indicates incomplete recognition.

Images